ALICE, ATLAS, CMS & LHCb joint workshop on DAQ Summary Thoughts ALICE, ATLAS, CMS & LHCb joint workshop on DAQ Château de Bossey 14 March 2013 Beat Jost / Cern

First of All Many thanks to the session organizers and the speakers who went through a lot of work gathering and assembling all the material Many thanks also to David, Frans and Pierre who did most of the nitty-gritty organisation. Thanks also to PH Department for taking over the charge for the venue and providing the coffee/tea breaks Last but not least thank YOU all for coming and making this event a success Personal Notes Indico Statistics Participant: ALICE 22 ATLAS 38 CMS 24 LHCb 10 Other 3 total 97 Not bad… Summary Thoughts, 14 March 2013

From David’s Introductory Remarks Summary Thoughts, 14 March 2013

Systems Today In general everyone is quite happy with their systems up to now Upgrades during LS1, i.e. for Run 2 are stimulated more by replacing old/obsolete hardware/technologies (PCI-X) than really new requirements New technologies, e.g. ≥10GbE/IB allow simplifications and rationalizations, especially for the event building networks (CMS, Atlas uniform network) Summary Thoughts, 14 March 2013

Systems Tomorrow Tomorrow  Run 3+ Atlas and CMS no major upgrades (prob) in the DAQ systems, besides the standard 5-year replacement cycle (maybe taking advantage of newer technologies) Alice and LHCb are foreseeing a major upgrade for this period Alice: ~2 orders of magnitude higher data rates Continuous TPC readout Full online reconstruction throwing away raw data (ambitious/daring) Making Online computing infrastructure a GP Alice computing centre used online and offline LHCb: eliminate hardware trigger in the long run (LLT as intermediate stage) ~40 times higher data rate No architectural changes foreseen, just application of modern technologies, if they work… R&D As usual higher rate to tape (somehow reoccurring theme) Summary Thoughts, 14 March 2013

Systems Next Week Major steps for Atlas and CMS Personal note Introduction of new trigger level (Track trigger) after high pT trigger Higher readout rates (500-1000 kHz) compared with 100 kHz today And again… higher rate to tape Personal note I would hope that the two experiments can come together and find a solution based on a common effort This is a unique opportunity Summary Thoughts, 14 March 2013

HLT Developments Demands on CPU power will always be increasing Higher multiplicity Do more in HLT to increase sensitivity/purity What can we do to match Moore’s law still helps Not in clock speed More cores per chip Memory is the bottle neck (amount and bandwidth) Forking (checkpointing) and COW help for some time (surely till LS2) One event one process (maybe some parallel algorithms) is still the most efficient Have to make sure that we use the capabilities of the CPUs to their maximum ‘Coprocessors’ (GPU, XEON-PHI)… Very sexy (fashion only?) Needs a lot of work to get the code running efficiently and producing identical results Use ‘non-physics’ time of LHC (deferred triggering) Summary Thoughts, 14 March 2013

Links and Boards Parallel buses are dead Already for a long time for data (performance) Now also for controls Fashion for boards nowadays xTCA Features: Support for high-power boards (cooling, power) Board-board communication paths (PtP, star/mesh topologies) Only used in trigger applications Very small form factor Links If GBT comes in time and works, will most likely be used for transferring data out of detector Lot of effort invested. Performance? Summary Thoughts, 14 March 2013

Other Topics Storage System Monitoring and management Disks get bigger and bigger Problem is to organize the data over the disks (RAID, ODS) Recovery from failures In general we will always be able to write out our data Maybe not for free but surely affordable System Monitoring and management Very wide field and myriad of approaches Often fashion driven Looks as if convergence will happen eventually, maybe,with some luck to some degree Summary Thoughts, 14 March 2013

Conclusion (1) I think the workshop was useful If only to ‘learn’ what others are doing Actually I learned a lot even of what happens in LHCb… Can we find common solutions to common problems? Yes and no… Seems to have some new motivation to do it in the young generation LS1: Share experience (e.g. on common tools) LS2: further common discussion about choices Some topics have clearly been suggested DQM/Histogramming Control Technology tracking and tools testing PON Dataflow Monitoring Summary Thoughts, 14 March 2013

Conclusion (2) Should we repeat it? Well… up to you! Maybe not every 2 weeks, but maybe in 12 months (well before the end of LS1) or in 18 months (just before beam)? Groups talking to each other between now and the next edition Either on specific topics as suggested during the workshop Or make a joint wrap-up brainstorming between a few people to propose the next steps I was missing the controls (DCS) community. Maybe a consequence of the title of the workshop. Summary Thoughts, 14 March 2013